Smart cameras’ built-in capacity makes for facial recognition fit

While smart cameras are designed for facial recognition systems, they are not limited to this specific function. Aside from facial recognition, they are used in tasks such as object detection, motion detection, audio detection, remote monitoring, analytics and insights, and their integration with smart home systems. How well-suited they are to face biometrics applications depends on the resolution and other characteristics of the images they capture, but also the capacity the cameras have to store and process images.

This article will focus on two of the leading smart camera models specifically tailored for facial recognition systems. Their memory usage when processing facial recognition software is considered, their potential capacity for storing facial recognition databases explored, and the most efficient methods for handling large volumes of real-time video feeds identified.

Facial recognition compares biometric templates, rather than raw image files. Templates used in facial recognition tend to be around 1kB, according to a post published by Paravision earlier this year. Algorithms for matching and related processing functions make up libraries that range significantly in size. A Rank One Computing explanation from 2019 points out issues with the way NIST measures model size, but estimates a range from the 40MB of its own model up to 4GB.

Wyze Cam v3 Pro

Smart home company Wyze offers a budget smart camera, Cam v3 Pro, equipped with a robust edge AI processor, capable of handling data for smart person detection alerts. The Wyze Cam v3 Pro is designed with an IP65 rating, making it suitable for outdoor applications, particularly within smart home systems. However, it is worth noting that this security surveillance smart camera leverages a better night vision resolution and improved facial recognition capabilities, according to the manufacturer.

The Wyze Cam v3 Pro uses edge intelligence for person detection, enabling it to deliver instant alerts to users. This represents a significant upgrade compared to its predecessor, which relied on cloud processing for facial recognition alerts. Wyze has introduced a software service combining edge processing and cloud capabilities. Through the Cam Plus subscription, users can upgrade their Wyze security camera by selecting various detection options in the settings options, including persons, vehicles, and pets.

Here’s how it works: The edge AI runs a preliminary live stream analysis and then forwards this data to the cloud AI to verify in better detail for even more accurate detection. This combination allows users to leverage faster processing and instant alerts of edge processing while also benefiting from the verification capabilities of cloud computing. In addition to intelligent detection, the Cam Plus software subscription also offers cloud recording functionality, storing videos in the cloud storage for up to 14 days. This contrasts with the native microSD card storage, which has limited capacity and offers longer retention than cloud-based recording.

Processing	Memory (Processing)	Storage (Database)
Edge	1GB	Up to 256GB microSD in exFAT format
Cloud	1GB	Up to 14 days of storage length

In the context of facial recognition smart cameras, memory serves as a volatile resource that provides temporary storage for data and programs. When the camera sensor captures a live video stream or an image, it temporarily holds this information in the memory for immediate processing. Subsequently, the facial recognition algorithms, situated at the edge, analyze the data in memory to identify and verify faces by comparing them to a database of known faces.

Conversely, storage is a non-volatile memory intended for storing data for long-term use, such as a database of images, videos, and metadata. In facial recognition smart cameras, storage is crucial in storing captured images and videos for later reference, analysis, or evidence collection. This functionality proves particularly critical in security and surveillance applications, where footage may need to be retained for an extended duration.

Vivotek Facial Recognition Camera

Earlier this year, Vivotek, a manufacturer specializing in IP surveillance embedded devices, released its FD9387 series camera with a smart motion detection feature. The standout element of this feature is its human detection capability. The camera module leverages a human silhouette database and uses an artificial neural network technology to recognize human appearances in video surveillance footage.

The FD9387-FR-v2 is an outdoor network camera that provides high-resolution video at 30 frames per second. Vivotek claims that that camera offers better nighttime surveillance because it utilizes Smart IR II technology, which optimizes IR intensity, reduces IR hotspots, and extends the IR effective range up to 50 meters. This surveillance camera is equipped to store up to 10,000 facial profiles in its internal memory at the edge and deliver facial recognition accuracy of up to 99 percent, as claimed by the company.

The camera operates with edge intelligence without needing middleware or FTP servers, simplifying the deployment process. Additionally, its synchronization of facial profiles with other cameras in the system makes it suitable for a multi-camera smart home surveillance system. In particular, Vivotek has also implemented a privacy mode to safeguard personal information, allowing users to control the sharing of facial recognition data.

Processing	Memory (Processing)	Storage (Database)
Edge	DDR4 2GB	1.     eMMC: 1GB 2.     microSD: up to 64GB

A 2GB memory capacity can significantly influence the efficiency of facial recognition operations by providing enough storage for crucial data involved in facial recognition processes, such as live video streams and images. This memory allocation accelerates processing at the edge of the system.

In contrast, the integrated embedded MultiMedia Card (eMMC) offers a default 1GB of storage, with the flexibility to expand it by inserting a microSD card with a capacity of up to 64GB. This expansion option plays a vital role in accommodating facial recognition databases, including profiles of identified individuals.

What should you consider?

The market offers a wide range of smart surveillance camera options that suit various consumer needs, ranging from industrial use cases to smart home systems. However, it’s important to determine which smart camera is best for your requirements and what specifications should be considered. Apart from the pivotal concerns regarding storage and memory, implementing facial recognition algorithms can take various forms.

A previous Biometric Update post explains some of these implementation types, including edge-based, cloud-native, and hybrid combinations. Taking the Wyze Cam v3 Pro as an example, it uses a hybrid approach, performing initial analysis on the device at the edge while delegating certain complex facial recognition tasks to the cloud. Moreover, the company offers users the choice between a native storage system via a microSD card and cloud storage. This flexibility is valuable for consumers deploying multiple smart cameras in far-edge locations, particularly in industrial environments.

However, for a smart home system, which typically requires no more than five cameras (for example), the Vivotek network camera, our second example, is a suitable choice. Regarding factors to consider, this camera relies on edge intelligence with database storage housed inside the camera’s memory. This edge-only facial recognition implementation is well-suited for smaller implementations in smart home systems while retaining the scalability to serve as a foundation for large surveillance infrastructure.

About the author

Abhishek Jadhav is a Master’s graduate in Electrical Engineering and a technology and science writer at EdgeIR,

Article Topics

biometrics  |  cameras  |  consumer electronics  |  facial recognition  |  video surveillance  |  Vivotek  |  Wyze